Skip to content

Lab: M11: deploy your AI app, then build your capstone

You'll need: your M4 setup (venv, key in .env, anthropic), plus fastapi + uvicorn, and Docker for Part B (optional, there's a fallback). Time: ~70 min + capstone • Work in your breakout pair.

Heads up: Part B needs Docker, which is a bigger install. If it fights you, do the no-Docker fallback in Step 7, you still finish with a real running web service. Errors are normal and safe.

This lab has two parts, then the capstone kickoff: - Part A: wrap your app in a FastAPI web API. - Part B: containerize it with Docker (or the uvicorn fallback).

flowchart LR
  Script["your AI code<br/>(a function)"] --> API["FastAPI<br/>/chat endpoint"]
  API --> Box["Docker image<br/>(app + dependencies)"]
  Box --> Run["runs the same anywhere<br/>key passed at run time"]

Part A: wrap it in a web API

Step 1: Install FastAPI and set up the folder

With your venv active: 

pip install fastapi "uvicorn[standard]"
Put app.py, requirements.txt, Dockerfile, .dockerignore (from solution/) in a folder with your M4 .env.

You should now see: Successfully installed fastapi-… uvicorn-…, and those files in the folder.

Step 2: Start the server

uvicorn app:app --reload
You should now see: Uvicorn running on http://127.0.0.1:8000. Your script is now a live web service. Leave it running.

Step 3: Try it in the browser

Open http://127.0.0.1:8000/docs. This is FastAPI's auto-generated test page. Expand POST /chat, click Try it out, enter {"message": "Say hi in one sentence"}, and Execute.

You should now see: a JSON response like {"reply": "Hi there! ..."}, and in the terminal a log line: POST /chat ok latency=… in_tokens=… out_tokens=…. You just called your AI over HTTP, and you're already monitoring latency and cost.

Step 4: Read the API

Open app.py. Find /health (a liveness check monitors hit), the ChatRequest/ChatResponse models (FastAPI validates these), and the logging line.

You should now see / say: "a request comes in → FastAPI validates it → my function calls the model → I log latency + tokens → the reply goes back as JSON." That's a web service.

Part B: containerize it

Step 5: Build the image

Make sure Docker is installed and running (see the Docker guide). From the app folder: 

docker build -t ai-app .
You should now see: build steps finishing with Successfully tagged ai-app (or naming to … ai-app). Your app + its dependencies are now one image.

Step 6: Run the container (key passed safely)

docker run -p 8000:8000 --env-file .env ai-app
You should now see: Uvicorn running on http://0.0.0.0:8000. Open http://127.0.0.1:8000/docs and call /chat again, same app, now running inside a container. Your key came from --env-file at run time; it is not inside the image.

Step 7: (Fallback if Docker won't cooperate)

No Docker? You already have a real service from Part A: uvicorn app:app --host 0.0.0.0 --port 8000. That's a legitimate deployment, containers just make it portable and reproducible.

You should now see: either a running container (Step 6) or a running uvicorn server. Either way, your AI app is deployed as a service.

Capstone kickoff

Step 8: Pick a track and scope it

Open CAPSTONE.md. Choose RAG, agent, or your idea, and write down the smallest version that runs. Use ../starters/app_starter.py, plug your M7/M8 RAG or M9 agent into answer_question().

You should now see: a one-paragraph plan and a running skeleton (uvicorn app_starter:app --reload → /docs). From here you build, then demo.

Stuck? The finished service is ../solution/app.py.

Your win

Your AI app runs as a real web service, wrapped in FastAPI, containerized with Docker, key kept safe, latency and cost logged, and your capstone is scoped and started.

Post it to the chat wins board: "POST /chat → my AI, served over HTTP from a Docker container. It runs for real now "

Take-home (optional)

Finish and demo your capstone (see CAPSTONE.md for the requirements: it runs, handles a basic failure gracefully, and you can explain how it works and how you'd secure it). Then present it, everyone claps. You're an AI engineer.